ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Contents lists available at ScienceDirect

ISA Transactions
journal homepage: www.elsevier.com/locate/isatrans

Linear control of the flywheel inverted pendulum

Manuel Olivares a, Pedro Albertos b,n
a
Departamento de Electrónica, Universidad Técnica Federico Santa María, Valparaíso, Chile
b
Instituto de Automática e Informática Industrial, Universitat Politècnica de València, València 46022, Spain

art ic l e i nf o a b s t r a c t

Article history: The flywheel inverted pendulum is an underactuated mechanical system with a nonlinear model but
Received 29 July 2013 admitting a linear approximation around the unstable equilibrium point in the upper position. Although
Received in revised form underactuated systems usually require nonlinear controllers, the easy tuning and understanding of linear
19 December 2013
controllers make them more attractive for designers and final users. In a recent paper, a simple PID
Accepted 24 December 2013
controller was proposed by the authors, leading to an internally unstable controlled plant. To achieve
This paper was recommended for
publication by Dr. A.B. Rad global stability, two options are developed here: first by introducing an internal stabilizing controller
and second by replacing the PID controller by an observer-based state feedback control. Simulation and
experimental results show the effectiveness of the design.
Keywords: & 2013 ISA. Published by Elsevier Ltd. All rights reserved.
Inverted pendulum
Underactuated systems
PID control design
Internal instability

1. Introduction
Underactuated mechanical systems received a lot of interest as
they appear in many practical applications such as robotic systems
(e.g. mobile robots, flexible-link robots, snake-type robots, and
walking robots), aerospace systems (e.g. aircraft, spacecraft, heli-
copters, rockets and satellites), or marine vehicles (e.g. surface
vessels and underwater vehicles). They are characterized by the
fact that there are more degrees of freedom than actuators, i.e.,
one or more degrees of freedom are unactuated [1] presenting
challenging control problems to solve operational inconveniences
with great interest from the theoretical point of view. Most of the
reported works on this kind of mechanical systems approach the
problem from a nonlinear perspective [2–5]. The linear approx-
imation around equilibrium points may not, in general, be con-
trollable and the feedback stabilization approach to transform the
plant into a linear one, in general, cannot be used. Therefore linear
control methods are not used to solve the feedback stabilization
problem, not even locally. In the same way, the tracking control
problem cannot be transformed into a linear control problem.
But linear control systems are very appealing by their simplicity
and easy tuning. The design procedure may have different steps
in order to consider different situations but, in any case, a clear
understanding of the design parameters is at hand. With this idea in
mind, in our previous paper [6], a PID controller was proposed to
control a flywheel inverted pendulum (FIP) [7]. Due to the under-
actuation, a derivative behavior appears at the plant output, so
the upper equilibrium position can be reached for any constant or
null input value, if the overall system is stable. The problem already
reported is that the PID solution is internally unstable.
The paper is organized as follows. First, to fix the problem, the
nonlinear model and its approximated linearization around the
unstable equilibrium point for this well-known mechanical device
are derived. Then, the design of a PID controller to stabilize the plant
and to compensate measurement disturbances is reviewed. Looking
at the FIP model, it presents an unstable open-loop pole and a zero
at the origin. So, even though the input/output behavior of the
controlled plant appears to be stable, its internal stability is not
achieved. To stabilize the internally unstable controlled plant, two
options are considered: first a new control loop is added, keeping
the global stability achieved by the initial design and allowing to
control the unstable internal variable. The second option is a state
feedback control, where cancelation is avoided. This results in a good
behavior but it requires full access to the state, thus a state
estimator/observer should be implemented. These results are
illustrated experimentally by the control of a laboratory prototype.
Some comments and future works are outlined in the last section,
where the improvements with respect to the previous paper are
discussed.

2. Flywheel inverted pendulum

n
Corresponding author. Tel.: þ 34 96 3879570.
E-mail addresses: manuel.olivares@usm.cl (M. Olivares), A review on the control of underactuated systems can be found
pedro@aii.upv.es (P. Albertos). in [1,8] (where the classical inverted pendulum mounted on a cart

0019-0578/$ - see front matter & 2013 ISA. Published by Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.isatra.2013.12.030

Please cite this article as: Olivares M, Albertos P. Linear control of the flywheel inverted pendulum. ISA Transactions (2014), http://dx.
doi.org/10.1016/j.isatra.2013.12.030i
2 M. Olivares, P. Albertos / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

is used as a benchmark of underactuated system), as well as in flywheel)

[9,10], where controllers have been designed by using linearization-

d ∂L ∂L ∂R
based or energy-based methods.  þ ¼ τi ; i ¼ ϕ; θ ð2Þ
dt ∂q_ i ∂qi ∂q_ i
Among the different approaches to control the FIP, a swinging-
up control at the unstable equilibrium point, without flywheel The total kinetic energy can be easily expressed as
angular velocity control, has been reported [11,12], a fuzzy control _ 2 þ I θ_ 2 þ m L2 ϕ
_ 2
_ ¼ 12 ½I p ϕ
Eðq; qÞ w w p ð3Þ
is reported in [13] and a linear full space state control design
using pole assignment has been deeply studied in [14]. As already where Ip and Iw are the inertia moments of the pendulum bar with
mentioned, a PID simple solution was proposed in [6]. respect to the fulcrum and the flywheel with respect to its rotation
axis, respectively, and Lp is the pendulum length (from the flywheel
2.1. The model axis to the fulcrum).
The potential energy is given by
A FIP consists of an inverted pendulum pivoting on a friction- _ ¼ mp gLc cos ϕ þ mw gLp cos ϕ
Vðq; qÞ ð4Þ
less point with a rotating mass on the top. It is an abstraction of
a biped robot, with an articulated/motorized joint, where the leg where Lc is the pendulum mass center distance to the fulcrum.
is represented by a bar and the moving body is abstracted as Thence, altogether, the Lagrangian is given by
a rotating motor. The reaction torque generated by this rotation _ 2 þ 1 I θ_ 2  α cos ϕ
_ ¼ 12 α1 ϕ
Lðq; qÞ 2 w 2
allows moving forward/backward the upper part of the pendulum.
A local sensor placed at the bottom of the pendulum provides α1 ¼ mw L2p þ Ip
a measurement of its inclination. A picture of such a device [7] is α2 ¼ ðmp Lc þmw Lp Þg ð5Þ
shown in Fig. 1(a) and a schematic diagram of the pendulum is
depicted in Fig. 1(b). If (2) is applied, taking into account that the generalized moment
A DC motor controlled by the armature voltage is moving of the system and the dissipative moments are given by (6), where
the inertia wheel. The main parameters to be considered are the ηϕ and ηθ are the friction factors and τ is the external torque
armature resistance (R) and the inductance (L), as well as the applied to the flywheel,
torque constant (M). ∂R
τϕ ¼ I w θ€ ; _
¼ ηϕ ϕ
∂q_ ϕ
∂R
2.1.1. Lagrangian formulation τθ ¼ τ  Iw ϕ€ ; ¼ ηθ θ_ ð6Þ
Defining by fmp ; I p ; ϕg the pendulum mass, its moment of ∂q_ θ
inertia with respect to the base and its angular position with then the nonlinear model of the FIP is expressed by the coupled
respect to the vertical axis, respectively; and by fmw ; I w ; θg the equations:
flywheel mass, its moment of inertia with respect to its center of
mass and its rotation angle, respectively. α1 ϕ€ þ Iw θ€ ¼ α2 sin ϕ  ηϕ ϕ_ ð7Þ
The Lagrangian is defined by (1), where E, V denote the kinetic
and potential energies, respectively, and q is the generalized € þ θ€ Þ ¼ τ  η θ_
I w ðϕ ð8Þ
θ
coordinates vector of the system, q ¼ ½ϕ θT .
The input torque τ ¼ Mia generated by the electric motor, where ia
_ ¼ Eðq; qÞ
Lðq; qÞ _ V ðq; qÞ
_ ð1Þ is the armature current, is obtained from the differential equation
of the armature electric circuit:
The system dynamics is derived from the Euler–Lagrange
equation (2), where R is the Rayleigh0 s dissipative function and τi dia
va ¼ Ra ia þ La þvr ð9Þ
are the moments applied to each coordinate (pendulum bar and dt

Fig. 1. Flywheel inverted pendulum. (a) Physical device and (b) schematic representation.

Please cite this article as: Olivares M, Albertos P. Linear control of the flywheel inverted pendulum. ISA Transactions (2014), http://dx.
doi.org/10.1016/j.isatra.2013.12.030i
 M. Olivares, P. Albertos / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 3

Note that vr ¼ M θ_ is the e.m.f. and va, the armature voltage, is the 3. Experimental set-up and PID control
externally manipulated variable.
A laboratory prototype [7] is shown in Fig. 1. Since there are
2.2. Linearized model instrumentation constraints (only the pendulum angular position
ϕðtÞ is measured in this equipment) an output feedback controller
To analyze the dynamic behavior of the FIP around the unstable is foreseen.
upper position equilibrium point ðϕ0 ¼ 0Þ, the above model equations
are linearized. It is worth noting that only (7) is nonlinear and easy to 3.1. Numerical data
linearize, leading to the following equations:
For this lab prototype, the FIP parameters are provided in
α1 ϕ€ þ I w θ€ ¼ α2 ϕ  ηϕ ϕ_ ð10Þ
Table 1.
Thus Eqs. (5) and (19) yield α1 ¼ 0:0669, α2 ¼ 1:5725 and γ ¼
€ þ θ€ Þ ¼ Mi  η θ_
I w ðϕ ð11Þ
a θ 532:0341, leading to the following state space representation:
2 3 2 3
La _i a þ Ra ia þM θ_ ¼ va ð12Þ 0 1 0 0
6 0:0045 7 6 7
z_ ¼ 4 23:5031 0 5z þ 4 0:1393 5u;
The angular position ϕ is measured by means of an infrared sensor
 23:5031 0  19:9513  621:6819
located at the bottom of the pendulum, providing a proportional
voltage y ¼ ½107:47 0 0z ð20Þ

vϕ ¼ K m ϕ ð13Þ
3.2. Model-based design
The DC motor armature voltage is obtained from a low power voltage
input u by means of a power amplifier, given by
Let us consider the model derived above. From (20), the transfer
va ¼  K a u ð14Þ function G(s) is
Assuming a state vector x ¼ ½ϕ ϕ _ θ θ_ i T , and considering the single V ϕ ðsÞ
a 14:9751s
control input u from (14) and the output y ¼ vϕ given by (13), GðsÞ ¼ ¼ ð21Þ
UðsÞ ðs þ 19:95Þðs þ 4:847Þðs  4:848Þ
the state space representation is
2 3 showing three poles, one of them unstable, and one zero at the origin,
0 1 0 0 0 2 3
0 considering vϕ ðtÞ as the measured output variable and u(t) the single
6 α2  η η M 7
ϕ
6 α1  I w α1  I w 0 θ
α1  I w 7 6 7
6 α1  I w
7 6 0 7 control input. Due to the zero at the origin and the positive pole,
6
x_ ¼ 6 0 0 0 1 0 7x þ 6 0 7
7 6
7u the system is open loop unstable and non-minimum phase.
6  α2 ηϕ 7 6 7
6 α 1  I w α 1  I w 0  ηθ β M β 7 6 0 7 The output feedback control loop to stabilize the pendulum
4 5 4 5
M  Ra  Ka in the upward position is represented in Fig. 2, where δðtÞ is
0 0 0 L
La La a
a measurement disturbance, initially assumed to be null, and
y ¼ ½K m 0 0 0 0x ð15Þ the reference r ϕ is zero corresponding to the upward pendulum
position.
where
A basic PID controller C 1 ðsÞ was proposed in [6] to stabilize the
1 1 output, cancelling the fast stable plant pole and designing the closed
β¼ þ ð16Þ
α1  I w I w loop polynomial Acl ðsÞ ¼ ðs þ sÞ2 for Tð0Þ ¼ 2, yielding K c1 ¼ 16:1,
Simplified model: A simplified model can be obtained if some T i1 ¼ 0:26 and T d1 ¼ 0:04, leading to the input/output closed-loop
parameters are considered negligible. In particular La, ηθ and ηϕ transfer function (23):


can be neglected ðLa ¼ ηθ ¼ ηϕ ¼ 0Þ. It also appears that the third 1
C 1 ðsÞ ¼ K c1 1 þ þ T d1 s ð22Þ
variable x3 ¼ θ, the flywheel angular position, does not affect any T i1 s
other variable (third column of the system matrix (15) is null).
Thus, if there is no interest on its evolution, it can be discarded. 9:6951ðs þ 4:847Þ2 9:6951
Then, from (12), the armature current ia can be obtained TðsÞ ¼  ð23Þ
ðs þ4:847Þ3 s þ 4:847
1
ia ¼ ðK a u þ M θ_ Þ ð17Þ Note that T(s) is stable, but Tð0Þ cannot be designed to unity in
Ra steady-state even though the controller has an integrator, due to
and the FIP can be represented by the third order model the controller-plant pole/zero cancelation.
2 3 2 3
0 1 0 0
6 α2 0 M2 7 6 MK a 7
z_ ¼ 6
4 α1 α1 Ra 75z þ 4 α1 Ra 5u
Table 1
 α2 FIP parameters.
α1 0  M γ  Kaγ
  Symbol Description Value
y ¼ Km 0 0 z ð18Þ
_ θ_ T and
where z ¼ ½ϕ ϕ
mp Pendulum mass 0:3046 kg
mw Motor and flywheel mass 0:2804 kg

 Ip Pendulum inertial moment 0:043 kg m2
M 1 1
γ¼ þ ð19Þ Iw Flywheel inertial moment 1:5  10  5 kg m2
Ra I w α 1 Lp Pendulum length 0:292 m
In the next sections, the external input voltage u(t) correspond- Lc Pendulum mass center 0:258 m
ing to the control input, the output voltage vϕ ðtÞ corresponding to M Motor constant 0:0375 Nm=A
the pendulum angular position, and the unmeasured state ωðtÞ ¼ θ_ ðtÞ Ra Armature resistor 4:7 Ω
Ka Amplifier gain 1:1685 V=V
corresponding to the flywheel angular velocity are considered, Km Pendulum position sensor gain 107:47 V=rad
although it seems obvious that only one variable can be controlled g Gravitational acceleration 9:8 m=s2
with one control action.

Please cite this article as: Olivares M, Albertos P. Linear control of the flywheel inverted pendulum. ISA Transactions (2014), http://dx.
doi.org/10.1016/j.isatra.2013.12.030i
4 M. Olivares, P. Albertos / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

The responses, using the simulated plant, are plotted in Fig. 3.
Starting with the initial condition vϕ ð0Þ ¼ 0:2, the output is stabilized
and reaches the reference r ϕ ¼ 0. A step measurement disturbance
δðtÞ ¼ 0:2 is applied at t¼1.5 s. The response is faster, reaching a
steady-state error around vϕ;1   0:4, but a constant flywheel
angular acceleration, that is, a constant increment on the control
signal is needed to compensate it leading to saturation. This
control cannot be applied to the FIP as the control signal saturates
at 10 V (around 0.8 s later), and with such a small measurement
disturbance the pendulum falls down.
Thence, as can be seen in the control sensitivity function
Su ðsÞ ¼ UðsÞ=Rϕ ðsÞ where an integrator appears, the controlled plant
is internally unstable
C 1 ðsÞ
Su ðsÞ ¼
1 þ C 1 ðsÞGðsÞ
0:64741ðs þ 19:95Þðs þ 4:847Þ ðs  4:848Þ 2 s þ αf
Su ðsÞ ¼
sðs þ 4:847Þ2
0:64741ðs þ 19:95Þðs  4:848Þ
 ð24Þ
s the fast stable plant pole, giving the closed loop transfer function
4. Internally stable design
The problem of the internal instability is due to the pole/zero
cancelation at the origin. To avoid this cancelation, the PID con-
troller can be modified, shifting the pole at the origin but in return
Fig. 2. Output feedback FIP control.
the response will be slower. An alternative to remove the internal
instability is by introducing an additional loop to control the plant
input. This will also allow to determine the steady-state value of
the wheel speed, required to balance the pendulum at the zero
angular position. In this way, the saturation will not be reached.
The additional loop is conceived to stabilize and force the control
action to be constant introducing a r ϕ ðtÞ pendulum position reference
update to compensate measurement disturbances, with the control
scheme shown in the block diagram of Fig. 4.
The combined control loops are equivalent to the one shown in
Fig. 5 that should be designed to keep the whole system stable,
as the control sensitivity function (24) has embedded C 1 ðsÞ and
G(s). In this case, a first order lag filter (25) has been designed to
stabilize Su(s):
Kf
FðsÞ ¼ ð25Þ
The filter parameters K f ¼  1:24 and αf ¼ 19:95 are designed
to get a stable closed loop polynomial Acl ðsÞ ¼ ðs þ εÞ2 , for ε equal to
HðsÞ ¼ UðsÞ=Ru ðsÞ. Note that H(s) is stable, non-minimum phase and
has unitary static gain because of the Su(s) integrator
FðsÞC 1 ðsÞGðsÞ
HðsÞ ¼
1 þ C 1 ðsÞðFðsÞ þGðsÞÞ
 4:1158ðs þ 19:95Þ2 ðs þ 4:847Þ4 ðs  4:848Þ
¼
ðs þ19:95Þ3 ðs þ 4:847Þ4
 4:1158ðs  4:848Þ
 ð26Þ
s þ 19:95
The main advantage of the proposed design is the easy tuning
and understanding of the parameters. Those of the PID are selected
to control the output. Those of F(s) are determined to stabilize the
system and as a slave controller sets the flywheel speed.
The system responses are plotted in Fig. 6. The response to an
initial condition different from zero is very fast. The control action
u(t) is driven to the required initial zero reference r u ðtÞ ¼ 0 at

PID Control with Kc1=16.1 Ti1=0.26 and Td1=0.04

0.2

0
rφ, δ(t), vφ(t)

−0.2

−0.4

−0.6

−0.8

−1
0 0.5 1 1.5 2 2.5 3
t[s]

10

5
u(t)

−5

−10

0 0.5 1 1.5 2 2.5 3

t[s]

Fig. 3. Simulated FIP with PID control: time responses.

Please cite this article as: Olivares M, Albertos P. Linear control of the flywheel inverted pendulum. ISA Transactions (2014), http://dx.
doi.org/10.1016/j.isatra.2013.12.030i
 M. Olivares, P. Albertos / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎
t¼ 1 s. The whole system is stable, although slower than before.
Nevertheless, if there is a constant measurement disturbance
δðtÞ ¼ 0:2 at t¼1.5 s, the system output vϕ ðtÞ now reaches the zero
angular position with internal stability, as the reference r ϕ ðtÞ is
updated by the filter, with steady-state error in the control action
u1  3:2 for r u ðtÞ ¼ 0. This error is compensated changing r u ðtÞ ¼
δ=Fð0Þ at t ¼2.5 s, where Fð0Þ o 0 is the static gain of the filter,
setting ru to stabilize the pendulum with minimum energy (zero
flywheel angular velocity).
4.1. State feedback stabilizing controller
Another alternative, without the drawbacks of the PID control
strategy shown in Fig. 3, is the use of a state feedback control
designed to assign poles similar to those in (23). Assuming full
access to the state space vector z(t) of the simplified model (18),
Fig. 4. Internal instability outer loop compensator.
Fig. 5. Control sensitivity function stabilization loop.
1
rφ(t), δ(t), vφ(t)
0.5
−0.5
0 0.5
10
5
u(t), ru(t)
−5
−10
0 0.5
Fig. 6. Measurement disturbance stable compensation.
Please cite this article as: Olivares M, Albertos P. Linear control of the flywheel inverted pendulum. ISA Transactions (2014), http://dx.
doi.org/10.1016/j.isatra.2013.12.030i
5
and using the numerical model (20), the state feedback control law
u ¼  K sf x can be computed using LQR for Q ¼ I 33 and R ¼1000
leading the state feedback vector gain K sf ¼ ½2:29  103 4:72
102 0:08. The state feedback control structure is shown in Fig. 7,
and the ideal response is shown in Fig. 8.
In this case, the response is internally stable, but there is no
control on the steady-state flywheel speed. Moreover, in general,
the state vector is not fully accessible.
4.2. State-based output feedback stabilizing controller
As the only measurement is the pendulum position ϕ, the state
space vector variables ϕ _ and θ_ should be estimated or computed.
If a flywheel angular velocity sensor is available, with a gain K ω ¼
vω =ω, only the output derivative should be computed. Then, a PD
controller can be used.
The equivalent control structure is depicted in Fig. 9, where the
controller C 2 ðsÞ in (27) implements the first two terms of the
control law, and a proportional controller Kp (28) implements
the flywheel angular velocity feedback. A r ω ðtÞ reference for this
velocity can be also added, as shown in Fig. 9
K sf 1 þ K sf 2 s
C 2 ðsÞ ¼ ¼ K c2 ð1 þ T d2 sÞ ð27Þ
Km
Fig. 7. State feedback loop.
PID control with filter, Kf=−1.24 and αf=19.95
1 1.5 2 2.5 3 3.5 4
t[s]
1 1.5 2 2.5 3 3.5 4
t[s]
6 M. Olivares, P. Albertos / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

SF control with Ksf1=2.29e+003 Ksf2=472 and Ksf3=0.0771

0.2

0.1

rφ, δ(t), vφ(t)

rφ
0
δ(t)
vφ(t)
−0.1

−0.2
0 0.5 1 1.5 2 2.5 3
t[s]

3
2
1
u(t), vω(t)

0
−1
−2
−3
u(t)
−4 vω(t)

0 0.5 1 1.5 2 2.5 3

t[s]

Fig. 8. Ideal state feedback control: time responses.

4.4. Experimental results

The control strategy shown in Fig. 4 has been implemented

Fig. 9. State feedback based FIP control loop.
K sf 3
Kp ¼
Kω
These parameters are calculated from Ksf giving K c2 ¼ 21:3, T d2 ¼
0:21 and KP ¼2.31, considering a flywheel angular velocity sensor
gain given by K ω ¼ 301
V s=rad.
The system responses are plotted in Fig. 10, for the same initial
condition and measurement disturbance applied to the PID with
filter control strategy shown in Fig. 6. Now, the closed loop is
internally stable, as there is no pole/zero cancelation at the origin,
so r ϕ is kept constant. Also, a steady-state error in the flywheel
angular velocity vω1  1:25 is obtained in response to a step
change r ω ðtÞ ¼ 2:5 at t¼2.5 s, applied to stabilize the pendulum
with less energy. An augmented system with an error integrator
should be arranged to avoid this steady-state error.
4.3. Observer-based output feedback stabilizing controller
The previous solution requires an extra sensor for the flywheel
angular velocity, in addition to the pendulum position sensor
and a PD controller, which will amplify the measurement noise, if
any. If only the output is measured and the derivative noise is not
acceptable, an observer can be designed to estimate the full
state, as one of the advantages of the state observer is its filtering
properties.
with the laboratory prototype [7]. The experimental results
are summarized in the following figures. In Fig. 11, the inverted
pendulum angular position is initially controlled with the manu-
facturer internal PID analogue controller, showing high fre-
quency oscillations in the control action and the output signals.
The manufacturer analogue PID control based strategy introduces
strong ripple in the control action due to the continuous changes
made around the setpoint to deal with the internal instability.
At t¼ 25 s, the pendulum is stabilized in the upper position
ð28Þ
with the proposed PID plus filter control strategy. Then, at t¼42 s,
a change in the control input reference ru(t) is introduced, and
3 s later the equilibrium position is recovered, with minor distur-
bances, although a small chattering appears in steady-state. Then
at t¼63 s, the internal analogue PID control is restored. In the
same experiment, the evolution of the control action is plotted
in Fig. 12. At t ¼ 42 s the control input reference ru(t) has been
changed to get the pendulum stabilized with minimum energy.
5. Conclusions
An underactuated system, the so-called flywheel inverted pen-
dulum, has been studied and a simplified model has been derived.
The parameters of the experimental rig, available at the laboratory,
allow to get a numerical model used to analyze the structure of the
plant, as well as to design linear controllers.
The control structure is developed in two steps. First an output
stabilizer PID controller is obtained following a traditional control
design. The appearance of internal instability demands for a second
control loop, which has also been designed by a simple control
design method.
This strategy has succeeded to meet two simultaneous objec-
tives with only one control action. That is, the stabilization of the

Please cite this article as: Olivares M, Albertos P. Linear control of the flywheel inverted pendulum. ISA Transactions (2014), http://dx.
doi.org/10.1016/j.isatra.2013.12.030i
 M. Olivares, P. Albertos / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 7

PD − P simultaneous control with Kc2=21.3, Td2=0.21 and Kp=2.31

0.2
rφ
0.1 δ(t)

rφ, δ(t), vφ(t)

vφ(t)
0

−0.1

−0.2
0 0.5 1 1.5 2 2.5 3 3.5 4
t[s]

10
u(t)
rω(t)
5
vω(t)
u(t),rω(t),vω(t)

−5

−10

0 0.5 1 1.5 2 2.5 3 3.5 4

t[s]

Fig. 10. State feedback based control: time responses.

Experimental results PID with filter Experimental results PID with filter

−4 10

−4.5
5

−5
0

−5.5

−5

−6

−10
−6.5
15 20 25 30 35 40 45 50 55 60 65 15 20 25 30 35 40 45 50 55 60 65

Fig. 11. Experimental filtered PID FIP position control. Fig. 12. Experimental filtered PID FIP control action.

pendulum in the unstable vertical position, with compensation of
small measurement disturbances, allowing the user to set the
flywheel angular velocity to achieve such stability with minimum
energy.
An alternative control structure could be implemented, based
on a state feedback controller, avoiding the pole/zero cancelation
problem leading to the internal instability, but it has not been
implemented because no flywheel angular velocity sensor is available
yet. This is a matter of further design (using an observer) and
experimentation.
Further refinements in the control structure and implementa-
tion are sought to reduce the noise in the control action. The
steady-state chattering seems to be unavoidable, up to now, when
the pendulum position is controlled with zero flywheel angular
velocity, due to nonlinearities such as the dead zone present in
that operating region. Also, the steady-state error in the control
action – flywheel angular velocity will be analyzed.
Acknowledgments
The authors want to thank the Universidad Técnica Federico Santa
María providing a grant to Professor Albertos as a research visitor.
References
[1] Spong MW. Underactuated mechanical systems, control problems in robotics
and automation. Springer, Berlin, Heidelberg: Springer-Verlag; 1997.
[2] Fantoni I, Lozano R. Non-linear control for underactuated mechanical systems.
Springer, Berlin, Heidelberg: Springer-Verlag; 2002.

Please cite this article as: Olivares M, Albertos P. Linear control of the flywheel inverted pendulum. ISA Transactions (2014), http://dx.
doi.org/10.1016/j.isatra.2013.12.030i
8 M. Olivares, P. Albertos / ISA Transactions ∎ (∎∎∎∎) ∎∎∎–∎∎∎

[3] Wang J-J. Stabilization and tracking control of x  z inverted pendulum with
sliding-mode control. ISA Trans 2012;51:763–70.
[4] Gomez-Estern F, Ortega R, Rubio F, Aracil J. Stabilization of a class of under-
actuated mechanical systems via total energy shaping. In: 40th IEEE con-
ference on decision and control. Orlando, FL, USA; 2001.
[5] Bettayeb M, Boussalem C, Mansouri R, Al-Saggaf U. Stabilization of an inverted
pendulum-cart system by fractional pi-state feedback. ISA Trans 2013. http:
//dx.doi.org/10.1016/j.isatra.2013.11.014.
[6] Olivares M, Albertos P. On the linear control of underactuated systems: the
flywheel inverted pendulum. In: 10th IEEE international conference on control
and automation (ICCA 2013). Hangzhou, China; 2013.
[7] E.D. Technologies. Operating manual of the inverted pendulum system, model
IP-NC; 2002.
[8] Yu H, Liu Y, Yang T. Tracking control of a pendulum-driven cart-pole under-
actuated system. In: IEEE international conference on systems, man and
cybernetics (SMC0 2007). Montreal, QC, Canada; 2007.
[9] Ortega R, Spong MW, Gomez-Estern F, Blankenstein G. Stabilization of a
class of underactuated mechanical systems via interconnection and damping
assignment. IEEE Trans Autom Control 2002;47(8):1218–33.
[10] Shiriaev A, Pogromsky A, Ludvigsen H, Egeland O. On global properties of
passivity-based control of an inverted pendulum. Int J Rob Nonlinear Control
2000;10:283–300.
[11] Kelly R, Campa R. Control basado en IDA-PBC del péndulo con rueda inercial:
análisis en formulación Lagrangiana. Revi Iberoam de Autom e Inf Ind (RIAI)
2005;2:36–42.
[12] Spong MW, Corke P, Lozano R. Nonlinear control of the reaction wheel
pendulum. Automatica 2001;37:1845–51.
[13] Ruan X, Wang Y. The modelling and control of flywheel inverted pendulum
system. In: 3rd IEEE international conference on computer science and information
technology. Chengdu, China; 2010.
[14] Block DJ, Aström KJ, Spong MW. The reaction wheel pendulum of synthesis
lectures on control and mechatronics, vol. 1. San Rafael, California: Morgan &
Claypool; 2007.

Please cite this article as: Olivares M, Albertos P. Linear control of the flywheel inverted pendulum. ISA Transactions (2014), http://dx.
doi.org/10.1016/j.isatra.2013.12.030i